Table 3.

Summary of linear regression models evaluated for human WNV incidence rates and WNV-positive corvid death ratios per census tract in Atlanta, 2001–2007.^a

					2000 U.S. Census
Response variable Modelb	Mean distance to CSO (m)	Mean distance to catch basin (m)	Mean tree cover (%)	Mean elevation (m)	Percent houses 1950s–1960s	Median household income (U.S. dollars)	No. dead corvids	Constant	AIC	ΔAIC	ωic
Human WNV incidence
1	−6E–5**	—	−0.0018	—	0.0276*	−1E–5*	−0.0362	3.60*	309.0	0.0	0.817
2	−6E–5**	5E–5	−0.0012	−0.0018	0.0287*	−2E–5*	−0.032	4.13*	313.1	4.1	0.105
3	−8E–5*	3E–5	—	—	—	—	−0.117*	2.65*	314.8	5.8	0.04
4	—	—	—	—	0.0381*	−2E–5*	—	3.58*	316.8	7.8	0.02
5	—	—	−0.020*	−0.0035	—	—	—	3.75*	316.9	7.9	0.02
∑ωi	9.7E–1c	0.0	0.02	0.0	0.94	9.2E–1	0.04
WNV-positive corvid death ratio
1	−6E–4*	—	0.194*	—	0.1095	1.0E–5*	—	4.26*	707.0	0	0.913
2	−6E–4*	2E–5	0.215*	−0.0159	0.1023	−6.0E–5	—	10.46*	711.7	4.7	0.09
3	—	—	—	—	0.186*	1.4E–4*	—	1.19	728.2	21.2	0.00
4	—	—	0.172*	−0.0017	—	—	—	4.83	728.4	21.4	0.00
5	−3E–4	−1E–5	—	—	—	—	—	11.61*	739.9	32.9	0.00
∑ωi	1.0	0.0	1.0	0.0	0.0	9.1E–1

^aResults show the parameter estimates and p-values for each predictor; different models are ordered from best to worst.

^bEach candidate model included 455 observations. Numbers indicate the parameter estimate for each variable included in a given model. Dependent variables (human WNV incidence rates and WNV-positive corvid death ratios) were log₁₀+1 transformed, whereas independent variables were estimated over each census tract.

^cAkaike weights, ω_i = exp(−1/2 ΔAIC) / ∑exp((−1/2 ΔAIC).

^cAfter including variables with p < 0.07 in the estimation of ∑ω_i. Without including marginally significant variables, the sum of Akaike weights drop to 0.04.

^*p < 0.05;

^**p < 0.07.